Hi Varun,

ideally this needs to be accounted for at the level of target function calculation, as explained here:
http://journals.iucr.org/d/issues/2013/02/00/dz5268/dz5268.pdf
Intensity statistics in the presence of translational noncrystallographic symmetry. Read RJ, Adams PD, McCoy AJ. Acta Cryst. D69, 176-183 (2013).

In reality, this is not yet implemented for refinement.

It's not given that the minimum of ML coincides with the minimum of R-factor, so in general minimization of ML does not guarantee drop in R-factor, as explained here:

ftp://ftp.ccp4.ac.uk/ccp4/newsletter/oct_99/08_ml.ps

Also, the goal of ML-based refinement is not to obtain lowest R (as explained in link above).

I wonder what happens if after LS refinement you switch back to ML: do R-factors remain around 27.3/29.8 or they increase?

Pavel

On 9/29/14 4:38 AM, Varun Bhaskar wrote:
Hi,

I collected a dataset for a binary complex (72kDa size) to 3.62A resolution which could be processed in P32 2 1 space group.  Data quality analysis with phenix.xtriage showed a presence of Pseudotranslational symmetry (Non-origin peak in the patterson with 60% intensity  at 0.33, -0.33, -0.013). I determined the structure with MR. I have 6 copies in ASU. I tried Zanuda for SG validation and this looks like the right space group.

For refinement when I use Maximum likelihood (ML) as the target function, my Rfactor/Rfree is around 31.3/32 but when I change it to Least square (LS) my Rfactor/Rfree drops to 27.3/29.8. The map quality after ML or LS refinement looks  very similar for most of the molecule, except for few places where the map from ML has slightly better features.

I read people use LS instead of ML only in the case of twinning but I am not sure if it could also be used in the case of PST.  Would the use of LS as the target function for refinement be right in this case? 


Thanks in advance

Best Regards
Varun